All Notebooks

This page contains a collection of notebooks that demonstrate how to use AG2. The notebooks are tagged with the topics they cover. For example, a notebook that demonstrates how to use function calling will be tagged with tool/function.

Enhanced Swarm Orchestration with AG2

Swarm Ochestration

orchestrationgroup chatswarm

Agent Chat with custom model loading

Define and load a custom model

integrationcustom model

`run` function examples with event processing

Using run and event processing

runevent-processingintegratefrontend

Managing Multiple MCP Servers with MCPClientSessionManager

Learn how to use MCPClientSessionManager to dynamically manage multiple MCP server sessions (stdio and SSE) within tool functions for flexible, resource-efficient agent workflows.

MCPsession managementmulti-serverMCPClientSessionManagerdynamic tools

Nested Chats for Tool Use in Conversational Chess

LLM-backed agents playing chess with each other using nested chats.

nested chattool/functionorchestration

Swarm Orchestration with AG2

Swarm Ochestration

orchestrationgroup chatswarm

From Dad Jokes To Sad Jokes: Function Calling with GPTAssistantAgent

Use tools in a GPTAssistantAgent Multi-Agent System by utilizing functions such as calling an API and writing to a file.

OpenAI Assistanttool/function

Run a standalone AssistantAgent

Run a standalone AssistantAgent, browsing the web using the BrowserUseTool

assistantagentrunbrowser-usewebscrapingfunction calling

RealtimeAgent in a Swarm Orchestration

Swarm Ochestration

orchestrationgroup chatswarm

OpenAI Responses API V2 Client - Complete Guide

OpenAI Responses API V2 Client with rich UnifiedResponse objects, stateful conversations, and agent integration

openairesponses-apiv2-clientmultimodalagents

WebSurferAgent with Firecrawl Integration

This notebook demonstrates how to use the WebSurferAgent with the Firecrawl tool for web scraping and crawling.

firecrawltoolscraping

Group Chat with Retrieval Augmented Generation

Implement and manage a multi-agent chat system using AG2, where AI assistants retrieve information, generate code, and interact collaboratively to solve complex tasks, especially in areas not covered by their training data.

group chatorchestrationRAG

Image Input/Output with OpenAI V2 Client and AG2 Agents

OpenAI V2 Client

openaiclients

Additional Model Request Fields with Amazon Bedrock in AG2

modelsbedrockamazon

RealtimeAgent with WebRTC connection

RealtimeAgent using websockets

realtimewebsockets

Using Neo4j's native GraphRAG SDK with AG2 agents for Question & Answering

Neo4j Native GraphRAG utilizes a knowledge graph and can be added as a capability to agents.

RAG

AutoBuild

Automatically build multi-agent system with AgentBuilder

autobuild

Auto Generated Agent Chat: Task Solving with Provided Tools as Functions

Register function calls using AssistantAgent and UserProxyAgent to execute python or shell code in customized ways. Demonstrating two ways of registering functions.

code generationtool/function

Conversational Workflows with MCP: A Marie Antoinette Take on The Eiffel Tower

MCP

Agent with memory using Mem0

Use Mem0 to create agents with memory.

mem0integrationmemory

RealtimeAgent with local websocket connection

RealtimeAgent using websockets

realtimewebsockets

Use LLamaIndexQueryEngine to query Markdown files

Use any LlamaIndex vector store as a Query Engine

agentsdocumentsRAGdocagentchromachromadbpinecone

ReasoningAgent - Advanced LLM Reasoning with Multiple Search Strategies

Use ReasoningAgent for o1 style reasoning in Agentic workflows with LLMs using AG2

reasoning agenttree of thoughts

DeepResearchAgent

DeepResearch Agent

agentsbrowser-useDeepResearchwebscrapingfunction calling

DuckDuckGo Search Tool

toolsDuckDuckGoweb-searchsearch

Agent Chat with Async Human Inputs

Async human inputs.

asynchuman

Use AG2 in Microsoft Fabric

agentschatmicrosoftfabricguides

Group Chat with Tools

agentstoolsgroupchatusersguides

Agent Chat with Multimodal Models: DALLE and GPT-4V

Multimodal agent chat with DALL-E and GPT-4v.

multimodalgpt-4v

Load the configuration including the response format

Agent Quickstart Examples

agentstoolsquickstartexamplesautogen

Currency Calculator: Task Solving with Provided Tools as Functions

Learn how to register function calls using AssistantAgent and UserProxyAgent.

tool/function

WebSurferAgent

WebSurfer Agent

agentsbrowser-usecrawl4aiwebscrapingfunction calling

Using a local Telemetry server to monitor a GraphRAG agent

FalkorDB GraphRAG utilises a knowledge graph and can be added as a capability to agents. Together with a swarm orchestration of agents is highly effective at providing a RAG capability.

RAGtool/functionswarm

Group Chat

Explore the utilization of large language models in automated group chat scenarios, where agents perform tasks collectively, demonstrating how they can be configured, interact with each other, and retrieve specific information from external resources.

orchestrationgroup chatcode generation

A Uniform interface to call different LLMs

Uniform interface to call different LLM.

integrationcustom model

Using RetrieveChat with Qdrant for Retrieve Augmented Code Generation and Question Answering

This notebook demonstrates the usage of QdrantRetrieveUserProxyAgent for RAG.

QdrantintegrationRAG

Solving Multiple Tasks in a Sequence of Chats with Different Conversable Agent Pairs

Use AG2 to solve a set of tasks with a sequence of chats.

orchestrationsequential chats

RealtimeAgent in a Swarm Orchestration

Swarm Ochestration

orchestrationgroup chatswarm

Auto Generated Agent Chat: Solving Tasks Requiring Web Info

Solve tasks requiring web info.

webcode generation

Wikipedia Search Tools

Perplexity Search Integration with AG2

toolsperplexityweb-searchsearch

Exponential Backoff and Retry Configuration with Amazon Bedrock in AG2

Configure exponential backoff and retry behavior for Amazon Bedrock API calls in AG2.

integrationbedrockamazonretryexponential backofferror handling

Wikipedia Agent

Search Wikipedia with WikipediaAgent

toolswikipediasearch

Adding Browsing Capabilities to AG2

toolsbrowser-usewebscrapingfunction calling

Small, Local Model (IBM Granite) Multi-Agent RAG

Optimizing Small, Local LLMs in Multi-Agent RAG Workflows using IBM Granite, Document Retrieval, Web Search, and Ollama

Small LLMsRAGWeb SearchIBM GraniteOllamaPlanningReflection

SocietyOfMindAgent

Explore the demonstration of the SocietyOfMindAgent in the AG2 library, which runs a group chat as an internal monologue, but appears to the external world as a single agent, offering a structured way to manage complex interactions among multiple agents and handle issues such as extracting responses from complex dialogues and dealing with context window constraints.

orchestrationnested chatgroup chat

Agent Chat with Multimodal Models: LLaVA

Leveraging multimodal models like llava.

multimodalllava

Anthropic Structured Outputs with AG2

Anthropic's structured outputs feature provides two powerful modes: JSON Outputs (Get validated JSON responses matching a specific schema) and Strict Tool Use (Guaranteed schema validation for tool inputs).

structured outputanthropic

Chat with OpenAI Assistant using function call in AG2: OSS Insights for Advanced GitHub Data Analysis

This Jupyter Notebook demonstrates how to leverage OSS Insight (Open Source Software Insight) for advanced GitHub data analysis by defining `Function calls` in AG2 for the OpenAI Assistant.

OpenAI Assistanttool/function

Use AG2 to Tune OpenAI Models

llmhyperparamatertuningopenaigptparameter tuning

Translating Video audio using Whisper and GPT-3.5-turbo

Use tools to extract and translate the transcript of a video file.

whispermultimodaltool/function

Agentic RAG workflow on tabular data from a PDF file

RAGgroupchat

Using ReliableTool to Generate Sub-Questions (Synchronous Version)

ReliableTool - Basic Example

tool callingtoolsreliable

Shell Tool with OpenAI Responses API

Use the shell tool with OpenAI's Responses API for filesystem diagnostics, extending capabilities with UNIX utilities, and running multi-step build and test flows.

responsesshellfilesystemautomationcliopenai

Creating Custom Events for Run Iteration

Creating and using custom events for step-by-step execution

custom-eventsrun-itereventsextensibility

Grok-powered SBOM Analysis with AG2

Using AG2 with Grok to perform SBOM analysis and security vulnerability scanning on Git repositories

grokopenai

CaptainAgent

Introducing CaptainAgent, a powerful agent that can manage and orchestrate other agents and tools to solve complex tasks.

autobuildCaptainAgent

Trip planning with a FalkorDB GraphRAG agent using GroupChat

FalkorDB GraphRAG utilises a knowledge graph and can be added as a capability to agents. Together with a swarm orchestration of agents is highly effective at providing a RAG capability.

RAGtool/functionswarm

Auto Generated Agent Chat: Teaching AI New Skills via Natural Language Interaction

Teach the agent news skills using natural language.

learningteaching

Firecrawl Tool Example

This notebook demonstrates how to use the FirecrawlTool to scrape, crawl, and map websites.

tool callingtoolsfirecrawl

Handling errors with the RunResponseProtocol

This notebook demonstrates handling errors with the RunResponseProtocol

errorsrunresponseprotocol

Web Scraping using Oxylabs Web Scraper API

Use AG2 to Web Scrape with Oxylabs

llmopenaiweb scrapingoxylabs

Google Drive Tools

agentstoolsgoogle drive

Preprocessing Chat History with `TransformMessages`

Preprocessing chat history with `TransformMessages`

long context handlingcapability

Agent with memory using Mem0

Use Mem0 to create agents with memory.

memory

Automatically Build Multi-agent System from Agent Library

Automatically build multi-agent system from agent library

autobuild

Config loader utility functions

utilityconfig

Adding Google Search Capability to AG2

Google Search

agentstoolssearchwebgooglereal-time search

Exploring GPT-5: Comparing Verbosity Levels in AG2 AgentChat

GPT5 verbosity example

gpt5verbosity

FSM - User can input speaker transition constraints

Explore the demonstration of the Finite State Machine implementation, which allows the user to input speaker transition constraints.

group chatfsmorchestration

Runtime Logging with AG2

Provide capabilities of runtime logging for debugging and performance analysis.

loggingdebugging

Iterating Over Agent Runs with AG2

Step-by-step execution for debugging and human-in-the-loop workflows

step-modedebugginghuman-in-the-loopruneventsrun_iterrun_group_chat_iter

AG2 + Gemini Thinking Config Variants

Google Gemini Thinking

googlegeminithinking

Auto Generated Agent Chat: Function Inception

Function Inception: Enable AG2 agents to update/remove functions during conversations.

function inceptiontool/function

Tavily Search Tool

Tavily Search Integration with AG2

toolstavilyweb-searchsearch

Using FalkorGraphRagCapability with agents for GraphRAG Question & Answering

RAGFalkorDB

Solving Multiple Tasks in a Sequence of Chats

Use conversational agents to solve a set of tasks with a sequence of chats.

orchestrationsequential chats

Using RetrieveChat Powered by PGVector for Retrieve Augmented Code Generation and Question Answering

Explore the use of AG2's RetrieveChat for tasks like code generation from docstrings, answering complex questions with human feedback, and exploiting features like Update Context, custom prompts, and few-shot learning.

PGVectorintegrationRAG

Using Neo4j's graph database with AG2 agents for Question & Answering

Neo4j GraphRAG utilises a knowledge graph and can be added as a capability to agents.

RAG

Structured output

LLM providers offer functionality for defining a structure of the messages generated by LLMs, AG2 enables this functionality by propagating a `response_format`, in the LLM configuration for your agents, to the underlying client.

structured output

Language Agent Tree Search

Language Agent Tree Search.

LATSsearchreasoningreflection

Assistants with Azure Cognitive Search and Azure Identity

This notebook demonstrates the use of Assistant Agents in conjunction with Azure Cognitive Search and Azure Identity

integrationRAGAzure IdentityAzure AI Search

Using RetrieveChat for Retrieve Augmented Code Generation and Question Answering

Explore the use of AG2's RetrieveChat for tasks like code generation from docstrings, answering complex questions with human feedback, and exploiting features like Update Context, custom prompts, and few-shot learning.

RAG

Using ReliableTool to Generate Sub-Questions (Synchronous Version)

ReliableTool - Group Chat Example

tool callingtoolsreliable

Veronica Core: Circuit Breaker for AG2 Agents

Use veronica-core's CircuitBreakerCapability to add per-agent circuit breakers and system-wide SAFE_MODE to AG2 multi-agent workflows.

toolsveronicacircuit-breakerlocal-llmreliability

Chat Context Dependency Injection

toolsdependency injectionfunction calling

Group Chat with Coder and Visualization Critic

Explore a group chat example using agents such as a coder and visualization agent.

group chatcode generation

Daytona Executor: Sandbox Code Execution for AG2 Agents

Use AG2 to execute code in secure, isolated Daytona cloud sandboxes

code-executiondaytonasandboxmulti-language

Using ReliableTool to Generate Sub-Questions (Synchronous Version)

ReliableTool - Google Search Example

tool callingtoolsreliablesearchgoogle

Discord, Slack, and Telegram messaging tools

Adding Browsing Capabilities to AG2

toolsbrowser-usewebscrapingfunction calling

V1/V2 Client Compatibility Demonstration

Client V1/V2 Compatibility

openaiclients

Using RetrieveChat Powered by MongoDB Atlas for Retrieve Augmented Code Generation and Question Answering

Explore the use of AG2's RetrieveChat for tasks like code generation from docstrings, answering complex questions with human feedback, and exploiting features like Update Context, custom prompts, and few-shot learning.

MongoDBintegrationRAG

Generate Dalle Images With Conversable Agents

Generate images with conversable agents.

capabilitymultimodal

Chatting with a teachable agent

Learn how to persist memories across chat sessions using the Teachability capability

teachabilitylearningRAGcapability

Using OpenAI’s Web Search Tool with AG2

Web Search Preview

toolssearchresponses api

Adding YouTube Search Capability to AG2

YouTube Search Integration with AG2

toolsyoutubevideosearch

Supercharging Web Crawling with Crawl4AI

toolsbrowser-usewebscrapingfunction calling

Using Guidance with AG2

Constrained responses via guidance.

guidanceintegrationJSON

Perform Research with Multi-Agent Group Chat

Perform research using a group chat with a number of specialized agents.

group chatplanningcode generation

Solving Complex Tasks with Nested Chats

Solve complex tasks with a chat nested as inner monologue.

nested chatreflectionreasoningorchestration

Perplexity Search Tool

Perplexity Search Integration with AG2

toolsperplexityweb-searchsearch

YepCode Executor: Serverless Code Execution for AG2 Agents

Use AG2 to execute code with YepCode

llmopenaiyepcode

OpenAI Assistants in AG2

Two-agent chat with OpenAI assistants.

OpenAI Assistant

Structured Outputs with Amazon Bedrock in AG2

Use Amazon Bedrock with AG2 to build applications that return structured data.

integrationcode generationbedrockamazonstructured outputpydantic

OptiGuide with Nested Chats in AG2

This is a nested chat re-implementation of OptiGuide which is an LLM-based supply chain optimization framework.

nested chathierarchical chatcode generationorchestration

RAG OpenAI Assistants in AG2

OpenAI Assistant with retrieval augmentation.

RAGOpenAI Assistant

Engaging Image Input/Output with OpenAI's Responses API in AG2

This notebook demonstrates how to do image input and image generate through a two agent chat with OpenAI Responses API and gpt-4o.

multimodalresponses

RemyxCodeExecutor: Agentic Code Exploration and Execution

AI-guided exploration and execution of research paper codebases using RemyxCodeExecutor

code executiondockerresearchremyx

AgentOptimizer: An Agentic Way to Train Your LLM Agent

AgentOptimizer is able to prompt LLMs to iteratively optimize function/skills of AutoGen agents according to the historical conversation and performance.

optimizationtool/function

Conversational Workflows with MCP: A French joke on a random Wikipedia article

MCP

Auto Generated Agent Chat: Collaborative Task Solving with Coding and Planning Agent

Use planning agent in a function call.

planningorchestrationnested chattool/function

Auto Generated Agent Chat: GPTAssistant with Code Interpreter

This Jupyter Notebook showcases the integration of the Code Interpreter tool which executes Python code dynamically within applications.

OpenAI Assistantcode interpreter

Interactive LLM Agent Dealing with Data Stream

Automated continual learning from new data.

streamingasynclearning

Agent Observability with OpenLIT

Use OpenLIT to easily monitor AI agents in production with OpenTelemetry.

integrationmonitoringobservabilitydebugging

Agent Tools and Run Method Examples

Agent Quickstart Examples

agentstoolsquickstartexamplesautogen

Currency Converter (Multi-Currency, Autonomous, API+SearchTool)

This notebook demonstrates a fully autonomous, multi-currency conversion system.

tool callingtoolscurrencygoogle search

Conversational Chess using non-OpenAI clients

LLM-backed agents playing chess with each other using nested chats.

nested chattool/functionorchestration

Tools with Dependency Injection

Tools Dependency Injection

toolsdependency injectionfunction calling

OpenAI Responses API with function call

This notebook demonstrates how to use tool calling with OpenAI's Responses API.

tool callingtoolsfunction callingresponses

(Legacy) Implement Swarm-style orchestration with GroupChat

orchestrationgroup chatstateflowswarm

Usage tracking with AG2

cost calculation

cost

OpenAI Responses API with Structured Output

This example demonstrates how to use Structured Output with OpenAI response API client.

structured outputresponsescode generationdebugging

RealtimeAgent in a Swarm Orchestration

Swarm Ochestration

orchestrationgroup chatswarmrealtime

RealtimeAgent with gemini client

RealtimeAgent with gemini client using websockets

realtimewebsocketsgemini

Using RetrieveChat Powered by Couchbase Capella for Retrieve Augmented Code Generation and Question Answering

Explore the use of AG2's RetrieveChat for tasks like code generation from docstrings, answering complex questions with human feedback, and exploiting features like Update Context, custom prompts, and few-shot learning.

RAG

Auto Generated Agent Chat: Using MathChat to Solve Math Problems

Using MathChat to Solve Math Problems

math

Cross-Framework LLM Tool Integration with AG2

toolslangchaincrewaipydanticai

Solving Complex Tasks with A Sequence of Nested Chats

Solve complex tasks with one or more sequence chats nested as inner monologue.

nested chatsequential chatsorchestration

Demonstrating the `AgentEval` framework using the task of solving math problems as an example

AgentEval: a multi-agent system for assessing utility of LLM-powered applications

eval

Writing a software application using function calls

Equip your agent with functions that can efficiently implement features into your software application.

code generationtool/functionfastapisoftware engineering

Task Solving with Provided Tools as Functions (Asynchronous Function Calls)

Learn how to implement both synchronous and asynchronous function calls using AssistantAgent and UserProxyAgent in AutoGen, with examples of their application in individual and group chat settings for task execution with language models.

tool/functionasync

Auto Generated Agent Chat: Collaborative Task Solving with Multiple Agents and Human Users

Involve multiple human users via function calls and nested chat.

humantool/function

Maris: Fine-Grained Policy-Guided Safeguards Demo

safeguardmarispolicy-guided

Web Scraping using Apify Tools

Scrapping web pages and summarizing the content using agents with tools.

webapifyintegrationtool/function

Mitigating Prompt hacking with JSON Mode in Autogen

Use JSON mode and Agent Descriptions to mitigate prompt manipulation and control speaker transition.

JSONdescriptionprompt hackinggroup chatorchestration

Making OpenAI Assistants Teachable

Teach OpenAI assistants.

teachabilitycapabilitylearningRAGOpenAI Assistant

Task Solving with Code Generation, Execution and Debugging

Use conversable language learning model agents to solve tasks and provide automatic feedback through a comprehensive example of writing, executing, and debugging Python code to compare stock price changes.

code generation

Use ChromaDBQueryEngine to query Markdown files

ChromaDB Query Engine for document queries

agentsdocumentsRAGdocagentchromachromadb

Group Chat with Customized Speaker Selection Method

Introduce group chat with customized speaker selection method.

orchestrationgroup chat

DeepSeek: Adding Browsing Capabilities to AG2

toolsbrowser-usewebscrapingfunction callingdeepseek

Cross-Framework LLM Tool for CaptainAgent

toolslangchaincrewai

Agent Tracking with AgentOps

Use AgentOps to simplify the development process and monitor your agents in production.

integrationmonitoringdebugging

Groupchat with Llamaindex agents

Integrate llamaindex agents with Autogen.

reactllamaindexintegrationgroup chatsoftware engineering

Auto Generated Agent Chat: Group Chat with GPTAssistantAgent

Use GPTAssistantAgent in group chat.

OpenAI Assistantgroup chat

Solving Multiple Tasks in a Sequence of Async Chats

Use conversational agents to solve a set of tasks with a sequence of async chats.

orchestrationasyncsequential chats

Structured output from json configuration

OpenAI offers a functionality for defining a structure of the messages generated by LLMs, AutoGen enables this functionality by propagating response_format passed to your agents to the underlying client.

structured output

Auto Generated Agent Chat: Task Solving with Code Generation, Execution, Debugging & Human Feedback

Code generation, execution, debugging and human feedback.

humancode generation

Websockets: Streaming input and output using websockets

Websockets facilitate real-time, bidirectional communication between web clients and servers, enhancing the responsiveness and interactivity of AG2-powered applications.

websocketsstreaming

Engaging with Multimodal Models: GPT-4V in AG2

Leveraging multimodal models through two different methodologies: MultimodalConversableAgent and VisionCapability.

multimodalgpt-4v

GPT-5.1 Apply Patch Tool: Complete Guide to Automated Code Editing

Apply Patch Tool Example for GPT-5.1

toolsapply_patchgpt-5.1gpt-5in-built-tool

Validating Agent Output With FunctionTarget: A Comprehensive Guide to Custom Validation in Multi-Agent Conversations

Group Chat FunctionTarget Examples

group chatfunctiontargetgroupchat

RealtimeAgent in a Swarm Orchestration using WebRTC

Swarm Ochestration

orchestrationgroup chatswarm

Complete DevOps Pipeline Orchestration with Apply Patch & Shell Tools

DevOps with OpenAI Apply Patch & Shell Tools

devopstoolsbuilt-in-toolopenairesponses

SearxNG Search Tool Walkthrough

This notebook demonstrates how to use the SearxNG Search Tool for real-time web searches using a privacy-friendly, open-source metasearch engine.

toolsearxngsearchweb

StateFlow: Build Workflows through State-Oriented Actions

StateFlow: Build workflows through state-oriented actions.

orchestrationgroup chatstateflowresearch

Grok example with live search

grokopenai

Use MongoDBQueryEngine to query Markdown files

Mongo DB Query Engine

agentsdocumentsRAGdocagentmongodbquery

Use AG2 to Tune ChatGPT

llmhyperparamatertuninggptparameter tuning

Conversational Workflows with MCP: A Shakespearean Take on arXiv Abstracts

MCP

RAG with DocAgent

Query documents and web pages with DocAgent

agentsdocumentsRAGdocagent

Use AG2 in Databricks with DBRX

Use Databricks DBRX and Foundation Model APIs to build AG2 applications backed by open-source LLMs.

integrationcode generationdbrxdatabricksopen sourcelakehousecustom modeldata intelligence

OpenTelemetry Tracing for AG2 Agents

Tracing AG2 agents with OpenTelemetry: instrument agents, LLM calls, and tool execution to observe multi-agent interactions.

observabilitytoolstracingopentelemetryjaegergrafana tempodatadog